Opengauss kernel analysis: query rewriting

This article is shared from Huawei cloud community 《

openGauss Kernel analysis （ Four ）： Query rewriting

》, author ： handsome and stern young fellow .

Query rewriting

SQL Language is rich and diverse , Very flexible , Different developers have different experiences , Handwritten SQL Sentences are also various , In addition, it can be automatically generated by tools .SQL Language is a descriptive language , The user of the database just describes the desired result , It doesn't care about how the data is obtained , Enter the SQL It is difficult for language to be expressed in an optimal form , It often implies some redundant information , This information can be mined to generate more efficient SQL sentence . Query rewriting is to put the user's input SQL Statement to a more efficient equivalent SQL, Query rewriting follows two basic principles .

•  Equivalence ： The original statement and the rewritten statement , The output is the same .

•  Efficiency ： The rewritten statement , It is more efficient in execution time and resource use than the original statement .

Query rewriting optimization can be based on the theory of relational algebra , For example, predicate push down 、 Sub query optimization, etc , It can also be optimized based on heuristic rules , for example Outer Join eliminate 、 Table connection elimination, etc . Query rewriting is a rule-based logic optimization .

At the code level , The schema of query rewriting is as follows ：

The following external connections are eliminated Outer2Inner— The process of query rewriting is analyzed by taking the conversion from external connection to internal connection as an example ： stay left outer join perhaps right outer join in , If there is a query condition that can logically contain IS NOT NULL, for example c1 > 0, You can convert a query to INNER JOIN, Thus, the intermediate result set produced by association processing is reduced

External connection elimination Outer2Inner

The following is an example to illustrate the differences between various multi table connection methods

create table t1(c1 int, c2 int);

create table t2(c1 int, c2 int);

insert into t1 values(1, 10);

insert into t1 values(2, 20);

insert into t1 values(3, 30);

insert into t2 values(1, 100);

insert into t2 values(3, 300);

insert into t2 values(5, 500);

Internal connection inner join： Returns the combination that both tables satisfy , It is equivalent to taking the intersection of two tables

SELECT * FROM t1 inner JOIN t2 ON t1.c1 = t2.c1;

Left connection  left outer join： Returns all rows in the left table , If the row in the left table does not match the row in the right table , Then the columns in the right table in the result return to null values

SELECT * FROM t1 Left OUTER JOIN t2 ON t1.c1 = t2.c1;

The right connection  right outer join： Returns all the rows in the right table , If the row in the right table does not match the row in the left table , The columns in the left table in the result return to null values

SELECT * FROM t1 right OUTER JOIN t2 ON t1.c1 = t2.c1;

Full connection  full join： Returns all rows in the left and right tables . When a row has no matching row in another table , Then the column in the other table returns to null , It is equivalent to taking two tables and combining them

SELECT * FROM t1 full JOIN t2 ON t1.c1 = t2.c1;

On the basis of the above experiments, add t2 Tabular where Conditions

left join and inner join The result is the same , This is because the query criteria contain WHERE t2.c2 >100 This condition ,t2 All mismatched tuples in the table are filtered out （ Include null value ）, Therefore, query transformation can be performed left-outer join -> inner join, Can effectively reduce t1 and t2 The result set produced by the association , Achieve the purpose of performance improvement .

stay openGauss Database system ,subquery_planner Will traverse the query tree rtable, See if there is RTE_JOIN Node of type exists , Set up hasOuterJoins Marker quantity , To enter into reduce_outer_joins Interface , When the external connection elimination conditions are met, the external connection elimination can be performed . reduce_outer_Joins Two actions are performed inside the function ,（1）reduce_outer_joins_pass1 Pre inspection , It's inspection jointree Whether there are external links in , And some information about reference tables , For action 2 Prepare for information collection , Key reference data structures reduce_outer_joins_state;（2）reduce_outer_joins_pass2 Really complete the elimination of external links .

void reduce_outer_joins(PlannerInfo* root)

{

reduce_outer_joins_state* state = NULL;

state = reduce_outer_joins_pass1((Node*)root->parse->jointree);

/* planner.c shouldn't have called me if no outer joins */

if (state == NULL || !state->contains_outer)

ereport(ERROR,

(errmodule(MOD_OPT),

errcode(ERRCODE_OPTIMIZER_INCONSISTENT_STATE),

(errmsg("so where are the outer joins?"))));

reduce_outer_joins_pass2((Node*)root->parse->jointree, state, root, NULL, NIL, NIL);

}

Using the analysis method of the previous issue , You can get the query tree memory structure （ Query tree Query In the structure targetList Store the target attribute semantic analysis results ,rtable Storage FROM The range table generated by clause ,jointree Of quals Field storage WHERE Expression tree for clause semantic analysis ）

contrast reduce_outer_joins Run the pre and post query tree ,jointree and rtable Medium jointype All by join_left Convert to join_inner, That is, the external connection has been changed to the internal connection

(gdb) p *((JoinExpr*)(parse->jointree->fromlist->head.data->ptr_value))

$1 = {type = T_JoinExpr, jointype = JOIN_INNER, isNatural = false, larg = 0x7fdfb345cd08, rarg = 0x7fdfb345e2e8, usingClause = 0x0, quals = 0x7fdfb2f0b8a8, alias = 0x0, rtindex = 3}

(gdb) p *(RangeTblEntry*)(parse->rtable->tail.data->ptr_value)

$2 = {type = T_RangeTblEntry, rtekind = RTE_JOIN, relname = 0x0, partAttrNum = 0x0, relid = 0, partitionOid = 0, isContainPartition = false, subpartitionOid = 0, isContainSubPartition = false,

refSynOid = 0, partid_list = 0x0, relkind = 0 '\000', isResultRel = false, tablesample = 0x0, timecapsule = 0x0, ispartrel = false, ignoreResetRelid = false, subquery = 0x0, security_barrier = false,

jointype = JOIN_INNER, …}

Click to follow , The first time to learn about Huawei's new cloud technology ~